Using machine learning to measure changes in invertebrate immunity

Abstract Object detection, a relatively recent development, allows for the identification of multiple objects in photographs. Numerous machine learning algorithms have been developed for object detection, particularly for small‐scale objects. Small‐object detection algorithms can be used to count a wide variety of objects from counts of individuals in drone imagery to cell counts on a microscopic slide. These algorithms include faster region‐based convolutional neural network (R‐CNN), You Only Look Once (YOLO) and single shot detector (SSD). Each has its own strengths and weaknesses. To measure immune response in animal populations to both biotic (e.g. disease exposure) and abiotic conditions (e.g. temperature), researchers rely on manual counting of cells, which is time‐intensive. For insects, this research centres on counting haemocytes. For controlling insect pests, species‐specific viruses often represent a major biocontrol agent. However, we still lack knowledge about how the immune system reacts to being challenged with a pathogen via haemocyte counts. Here, we use machine learning, which automatically counts haemocytes, to compare detection algorithms. To perform counts, we trained Faster R‐CNN, YOLO and SSD models on 398 photographs and validated training on 114 photographs. Of the algorithms compared, YOLOv8 was the most accurate and the quickest to train. Counts performed on YOLOv8 tended to be fairly accurate and matched closely with haemocyte counts done by hand. YOLOv5 followed closely behind in accuracy with SSD and Faster‐RCNN not far behind. Counting time was reduced from 60 s per photograph when counting by hand to less than a second using YOLOv8's detection capabilities. Practical Implication . Object detection methods such as these drastically speed up the process of measuring immune response and open the door for further research whether it be on biocontrol methods in insects or other questions of biological interest. In general, small object detection allows for quick processing of otherwise difficult data and can apply to multiple fields including use in drone imagery, counting cells and detecting crop yield from photographs.